在煤层超前探测反演计算中，电磁数据具有非线性和不适定性，导致反演的计算量较大，严重影响了超前探测的效率。人工神经网络算法可以提高反演计算的速度和准确度，但传统神经网络算法对初始模型依赖较高且易陷入局部极小值。因此，引入深度神经网络（Deep Neural Network，DNN）来解决上述问题，实现了对煤层地质信息精确、快速的预测，具有一定的参考价值和意义。

本文提出了一种基于DNN的煤层超前探测反演算法。首先在正演计算中采用平面分层介质中的并矢格林函数法，针对传统地电模型建立方法会产生冗余、异常数据的问题，引入一种带约束条件的地电模型建立方法，并通过正演计算得到数据样本，利用其对网络进行训练。其次，为提高算法的非线性计算能力，搭建十层全连接的DNN网络结构并对其优化方法进行改进，通过对大量数据样本的训练找到反演计算中的复杂映射关系，对煤层结构进行超前探测时只需输入电场分量值等参数，即可得到相应的地电模型的分层介质参数，从而实现煤层地质的重构。在数值仿真中，分别对四层、七层地电模型的电导率、厚度参数进行预测，结果表明，DNN在计算精度高于BP（Back Propagation）神经网络的基础上，计算速度相较于BP神经网络提高了10-12倍。此外，为提高算法的泛化能力，提出基于多频点-多接收点的融合型数据训练DNN的方法，数值仿真表明，相较于单频点、单接收点的数据，通过融合型数据训练DNN会有效提高网络的泛化性。

最后将DNN算法应用于实测数据反演计算中，数值仿真结果表明，相较于BP神经网络，DNN预测结果精度高9.46%，且计算速度快12倍。本文采用DNN进行煤层超前探测反演计算可以有效提高准确度和计算速度，为实际情况中煤层地质结构的反演研究提供了一定的理论参考。

In the calculation of coal seam advanced detection inversion, the electromagnetic data has nonlinearity and unsuitable qualification, resulting in a large amount of calculation of inversion, which seriously affects the efficiency of advanced detection. Artificial neural network algorithms can improve the speed and accuracy of inversion calculations, but traditional neural network algorithms have high dependence on the initial model and are prone to falling into local minimums. Therefore, the introduction of Deep Neural Network (DNN) to solve the above problems and realize the accurate and rapid prediction of coal seam geological information has certain reference value and significance.

In this thesis, a DNN-based coal seam advanced detection inversion algorithm is proposed. Firstly, the parallel vector green function method in planar hierarchical medium is adopted in the forward calculation, and a geoelectric model establishment method with constraints is introduced to solve the problem that the traditional geoelectric model establishment method will produce redundant and abnormal data, and the data samples are obtained through the forward calculation and the network is trained. Secondly, in order to improve the nonlinear computing ability of the algorithm, build a ten-layer fully connected DNN network structure and improve its optimization method, find the complex mapping relationship in the inversion calculation through the training of a large number of data samples, and only need to input the electric field component value and other parameters when the coal seam structure is advanced detection, and the corresponding layered medium parameters of the geoelectric model can be obtained, so as to realize the reconstruction of coal seam geology. In the numerical simulation, the conductivity and thickness parameters of the four-layer and seven-layer geoelectric models are predicted, and the results show that the calculation speed of DNN is 10-12 times higher than that of BP neural network on the basis of higher calculation accuracy than BP neural network. In addition, in order to improve the generalization ability of the algorithm, a method of training DNN based on multi-frequency point-multi-receiving point fusion data is proposed, and numerical simulation shows that compared with the data of single frequency point and single receiving point, training DNN with fusion data will effectively improve the generalization of the network.

Finally, the DNN algorithm is applied to the inversion calculation of the measured data, and the numerical simulation results show that compared with the BP neural network, the accuracy of the DNN prediction result is 9.46% higher and the calculation speed is 12 times faster. In this thesis, DNN is used to calculate the inversion of coal seam advanced detection, which can effectively improve the accuracy and calculation speed, and provide a certain theoretical reference for the inversion study of coal seam geological structure in actual conditions.

[1]王力.地质环境超前探测与成像技术研究综述[J].煤炭技术,2021,40(06):83-85.

[2]董书宁,刘其声,王皓等.煤层底板水害超前区域治理理论框架与关键技术[J].煤田地质与勘探,2023,51(01):185-195.

[3]张平松,李圣,邱实,等.巷道快速智能掘进超前探测技术与发展[J].煤炭学报,2021,46(07):2158-2173.

[4]岳建华,杨海燕,冉华赓.矿井电法勘探研究现状与发展趋势[J].煤田地质与勘探,2023,51(01):259-276.

[5]张幼振,范涛,阚志涛,等.煤矿巷道掘进超前钻探技术应用与发展[J].煤田地质与勘探,2021,49(05):286-293.

[6]程久龙,赵家宏,董毅等.基于LBA-BP的矿井瞬变电磁法岩层富水性的定量预测研究[J].煤炭学报,2020,45(01):330-337.DOI:10.13225/j.cnki.jccs.YG19.1508.

[7]Pholsena K, Pan L, Zheng Z. Mode decomposition based deep learning model for multi-section traffic prediction[J]. World Wide Web, 2020, 23: 2513-2527.

[8]Jena R, Pradhan B, Al-Amri A, et al. Earthquake probability assessment for the Indian subcontinent using deep learning[J]. Sensors, 2020, 20(16): 4369.

[9]Hasan N I, Bhattacharjee A. Deep learning approach to cardiovascular disease classification employing modified ECG signal from empirical mode decomposition[J]. Biomedical signal processing and control, 2019, 52: 128-140.

[10]Wang Y, Chen Y, Bhuiyan M Z A, et al. Gait-based human identification using acoustic sensor and deep neural network[J]. Future Generation Computer Systems, 2018, 86: 1228-1237.

[11]Zu S, Cao J, Qu S, et al. Iterative deblending for simultaneous source data using the deep neural networkIterative deblending with DNN[J]. Geophysics, 2020, 85(2): V131-V141.

[12]Wei Z, Liu D, Chen X. Dominant-current deep learning scheme for electrical impedance tomography[J]. IEEE Transactions on Biomedical Engineering, 2019, 66(9): 2546-2555.

[13]Aggarwal H K, Mani M P, Jacob M. MoDL: Model-based deep learning architecture for inverse problems[J]. IEEE transactions on medical imaging, 2018, 38(2): 394-405.

[14]Wang B, Zhang N, Lu W, et al. Deep-learning-based seismic data interpolation: A preliminary result[J]. Geophysics, 2019, 84(1): V11-V20.

[15]Puzyrev V. Deep learning electromagnetic inversion with convolutional neural networks[J]. Geophysical Journal International, 2019, 218(2): 817-832.

[16]Holm-Jensen T, Hansen T M. Linear waveform tomography inversion using machine learning algorithms[J]. Mathematical Geosciences, 2020, 52(1): 31-51.

[17]HINTONGE,SALAKHUTDINOVRR.Reducing the Dimensionality of Data with Neural Networks[J].Science, 2006,313(5786):504.

[18]Zhang J, Zheng Y, Qi D, et al. DNN-based prediction model for spatio-temporal data[C]//Proceedings of the 24th ACM SIGSPATIAL international conference on advances in geographic information systems. 2016: 1-4.

[19]Du M, Liu N, Song Q, et al. Towards explanation of dnn-based prediction with guided feature inversion[C]//Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2018: 1358-1367.

[20]Lorenzo-Trueba J, Henter G E, Takaki S, et al. Investigating different representations for modeling and controlling multiple emotions in DNN-based speech synthesis[J]. Speech Communication, 2018, 99: 135-143.

[21]Ravindran R, Santora M J, Jamali M M. Multi-object detection and tracking, based on DNN, for autonomous vehicles: A review[J]. IEEE Sensors Journal, 2020, 21(5): 5668-5677.

[22]Zhang Y, Xie Y, Zhang Y, et al. The adoption of deep neural network (DNN) to the prediction of soil liquefaction based on shear wave velocity[J]. Bulletin of Engineering Geology and the Environment, 2021, 80: 5053-5060.

[23]Wen T, Xie G, Cao Y, et al. A DNN-based channel model for network planning in train control systems[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 23(3): 2392-2399.

[24]李刚,潘和平,王智等.回线源瞬变电磁法一维正演并行算法[J].地质科技情报,2017,36(03):299-305.

[25]冯凯,秦策.大地电磁(MT)自适应有限元各向异性正演[J].吉林大学学报(地球科学版),2020,50(06):1887-1896.

[26]郭嵩巍,郑凯,刘小畔等.基于全区视电阻率瞬变电磁一维正演中滤波系数的选取[J].科学技术与工程,2021,21(03):911-917.

[27]王天意,侯征,杨进等.基于迭代有限元法的大地电磁二维正演[J/OL].地球物理学进展:1-9[2023-03-09].

[28]范涛,鲁晶津,王冰纯,刘磊,赵兆.瞬变电磁虚拟波场反演法在井下超前探测中的应用[J].煤炭科学技术,2017,45(10):8-15+47.

[29]Godio A, Santilano A. On the optimization of electromagnetic geophysical data: Application of the PSO algorithm[J]. Journal of Applied Geophysics, 2018, 148: 163-174.

[30]赵涵,景旭,李貅等.多辐射场源地空瞬变电磁一维反演方法研究[J].物探与化探,2019,43(01):132-142.

[31]Di Q, Xue G, Yin C, et al. New methods of controlled-source electromagnetic detection in China[J]. Science China Earth Sciences, 2020, 63: 1268-1277.

[32]周君君,胡祥云,肖调杰,白宁波.大地电磁一维正则化反演算法研究[J].煤炭科学技术,2021,49(08):174-180.

[33]徐正玉,付志红,廖先等.基于小回线瞬变电磁法的病害体精细探测应用分析[J].地球物理学进展,2022,37(02):892-901.

[34]韩晓冰,余思淼,梁冰洋等.基于DNN的矿井超前探测反演方法研究[J].煤炭技术,2023,42(03):117-121.

[35]Jiang F B, Dai Q W, Dong L. Nonlinear inversion of electrical resistivity imaging using pruning Bayesian neural networks[J]. Applied Geophysics, 2016, 13(2): 267-278.

[36]Kim Y, Nakata N. Geophysical inversion versus machine learning in inverse problems[J]. The Leading Edge, 2018, 37(12): 894-901.

[37]王鹤,刘梦琳,席振铢等.基于遗传神经网络的大地电磁反演[J].地球物理学报,2018,61(04):1563-1575.

[38]Moghadas D. One-dimensional deep learning inversion of electromagnetic induction data using convolutional neural network[J]. Geophysical Journal International, 2020, 222(1): 247-259.

[39]Puzyrev V, Swidinsky A. Inversion of 1D frequency-and time-domain electromagnetic data with convolutional neural networks[J]. Computers & geosciences, 2021, 149: 104681.

[40]范涛,薛国强,李萍等.瞬变电磁长短时记忆网络深度学习实时反演方法[J].地球物理学报,2022,65(09):3650-3663.

[41]Wu Y, Feng J. Development and application of artificial neural network[J]. Wireless Personal Communications, 2018, 102: 1645-1656.

[42]Zhang L, Wang F, Sun T, et al. A constrained optimization method based on BP neural network[J]. Neural Computing and Applications, 2018, 29: 413-421.

[43]朱聪聪,朱国维,张庆朝.BP神经网络在高密度电法反演中的改进与应用[J].煤炭技术,2017,36(12):90-92.

[44]Ye Z, Kim M K. Predicting electricity consumption in a building using an optimized back-propagation and Levenberg–Marquardt back-propagation neural network: Case study of a shopping mall in China[J]. Sustainable Cities and Society, 2018, 42: 176-183.

[45]Aldahdooh A, Hamidouche W, Fezza S A, et al. Adversarial example detection for DNN models: A review and experimental comparison[J]. Artificial Intelligence Review, 2022, 55(6): 4403-4462.

[46]廖晓龙,张志厚,范祥泰,路润琪,姚禹,曹云勇,徐正宣.基于改进粒子群深度神经网络的频率域航空电磁反演[J].中南大学学报(自然科学版),2020,51(08):2162-2173.

[47]Sze V, Chen Y H, Yang T J, et al. Efficient processing of deep neural networks: A tutorial and survey[J]. Proceedings of the IEEE, 2017, 105(12): 2295-2329.

[48]尹成友,于涛,唐伟.球形分层媒质中并矢格林函数的求解Ⅰ:并矢格林函数的推导[J].微波学报,2014,30(03):9-14+27.

[49]赵运进,田锰,黄勇刚等.基于有限元法的光子并矢格林函数重整化及其在自发辐射率和能级移动研究中的应用[J].物理学报,2018,67(19):157-167.

[50]魏宝君,曹景强,任广强等.用柱状成层各向异性介质的并矢Green函数模拟偏心对多分量感应测井响应的影响[J].地球物理学报,2017,60(03):1234-1247.

[51]Liu H, Long Z, Han F, et al. Frequency-domain reverse-time migration of ground penetrating radar based on layered medium green's functions[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2018, 11(8): 2957-2965.

[52]Faryad M, Lakhtakia A. Infinite-space dyadic Green functions in electromagnetism[M]. Morgan & Claypool Publishers, 2018.

[53]Jang H , Kim H J . Mapping deep-sea hydrothermal deposits with an in-loop transient electromagnetic method:Insights from 1D forward and inverse modeling[J]. Journal of Applied Geophysics, 2015, 123:170-176.

[54]Gan H H, Dai Q I, Xia T, et al. Hybridization numerical Greenes function of anisotropic inhomogeneous media with surface integral equation[J]. Microwave and Optical Technology Letters, 2017, 59(7): 1781-1786.

[55]梁冰洋. 半航空瞬变电磁探测和颅骨光声成像技术研究[D].厦门大学,2019.

[56]Yu S, Ma J. Deep learning for geophysics: Current and future trends[J]. Reviews of Geophysics, 2021, 59(3): e2021RG000742.

[57]Chang Z, Zhang Y, Chen W. Electricity price prediction based on hybrid model of adam optimized LSTM neural network and wavelet transform[J]. Energy, 2019, 187: 115804.

[58]孟当当. 神东布尔台矿深部煤层顶板水地面瞬变电磁探查试验研究[D].安徽理工大学,2021.